Synthetic Data Generation for Autonomous Driving
My dissertation was developed within the THEIA Innovation project from BOSCH.
This work contributes with the process of decision-making for overtaking scenarios. A new dataset was created using the CARLA simulator and a knowledge base was created containing characteristics and factors important for overtaking, such as driver behavior and the sequence of actions that ensure safety and legality during the overtaking maneuver. This knowledge was coded as a set of rules that were added to the synthetic dataset.
Statistical analysis and machine learning methods were then applied with two purposes: (1) to highlight the most important features involved in successful and unsuccessful overtaking scenarios, and (2) to support car decisions by predicting if a sequence of actions will result in a successful or unsuccessful overtaking maneuver.